JMIR Publications

ABSTRACT

Background:

Suicide remains a leading cause of death among young people globally, with those in low- and middle-income countries bear a particularly high burden. In China, suicide risk among young adults is rising, yet there are limited targeted effort to understand the sociocultural and psychological correlates of suicide ideation and attempts in this population. Importantly, the risk and protective factors of suicide may differ based on cultural context, and advanced analytical techniques may help investigate a range of factors.

Objective:

This study aimed to explore key correlates of lifetime suicide ideation and suicide attempts among Chinese young adults using a machine learning approach applied to a large, cross-sectional dataset.

Methods:

We analyzed a cross-sectional survey data across young adults (N = 94,930) in 63 universities in Jilin, China. A total of 158 candidate variables were used to identify key correlates of suicide ideation and attempts, including sociodemographic variables, mental health symptoms and diagnoses, trauma exposure, gender identity and sexual orientation, help-seeking attitudes, addictive behaviors, health behaviors, etc.). We trained and evaluated models using the Extreme Gradient Boosting (XGBoost) algorithm and utilized the Shapley Additive Explanations (SHAP) method to assess feature importance.

Results:

The prevalence of lifetime suicide ideation was 28.2% and suicide attempt was 2.6%. The XGBoost models showed strong discriminative ability, with median area under the receiver operating characteristic curve (auROC) values of 0.87 (IQR = .003, range = .87-.88) for suicidal ideation and 0.89 for suicide attempts (IQR = .02, range = .87-.91). Using SHAP values, the most critical correlates of both outcomes included non-suicidal self-injury (NSSI), childhood emotional abuse, reluctance to seek help for suicidal ideation, trauma symptoms, and depression. Additional correlates included obsessive-compulsive symptoms and nonconformity to traditional gender roles.

Conclusions:

Machine learning-based models identified a range of psychosocial and behavioral correlates of suicidality in a large sample of Chinese young adults. Findings underscore the relevance of NSSI, childhood emotional abuse, and stigma-related help-seeking behavior in understanding suicide risk in this cultural context. While the cross-sectional nature of the data limits causal inference, results provide a foundation for hypothesis generation and future longitudinal research and intervention development. These insights may inform culturally responsive suicide prevention strategies in Chinese youth and global efforts to understand and address suicide risk among youth in LMIC settings. Clinical Trial: No clinical trial was involved in this study.